HTTP/1.1 200 OK Date: Tue, 09 Apr 2002 06:39:03 GMT Server: Apache/1.3.20 (Unix) Last-Modified: Fri, 02 Feb 1996 23:00:00 GMT ETag: "304d7b-ca53-31129770" Accept-Ranges: bytes Content-Length: 51795 Connection: close Content-Type: text/plain PPP Working Group A. Valencia INTERNET DRAFT M. Littlewood Category: Experimental Draft T. Kolar Title: draft-ietf-pppext-l2f-00.txt Cisco Systems Date: February 1996 Level Two Forwarding (Protocol) "L2F" Status of this Memo This memo is an experimental draft. It describes an implemented pro- tocol, but one which is expected to change significantly as a result of review in the working group. Abstract Virtual dial-up allows many separate and autonomous protocol domains to share common access infrastructure including modems, Access Servers, and ISDN routers. Previous RFCs have specified protocols for supporting IP dial-up via SLIP and multiprotocol dial-up via PPP. This document describes the Level Two Forwarding protocol (L2F) which per- mits the tunneling of the link level (i.e., HDLC, async HDLC, or SLIP frames) of higher level protocols. Using such tunnels, it is possi- ble to divorce the location of the initial dial-up server from the location at which the dial-up protocol connection is terminated and access to the network provided. Table of Contents 1.0 Introduction 2 1.1 Conventions 2.0 Problem Space Overview 3 2.1 Initial Assumptions 2.2 Topology 2.3 Virtual dial-up Service - a walk-though 4 3.0 Service Model Issues 6 3.1 Security 3.2 Address allocation 3.3 Authentication 7 3.4 Accounting 4.0 Protocol Definition 8 4.1 Encapsulation within L2F 8 4.1.1 Encapsulation of PPP within L2F 4.1.2 Encapsulation of SLIP within L2F 9 4.2 L2F Packet Format 4.2.1 Overall Packet Format Valencia expires August 1996 [Page 1] RFC DRAFT January 1996 4.2.2 Packet Header 10 4.2.3 Version field 4.2.4 Protocol field 4.2.5 Sequence Number 11 4.2.6 Packet Multiplex ID 4.2.7 Client ID 4.2.8 Length 12 4.2.9 Packet Checksum 4.2.10 Payload Offset 4.2.11 Packet Key 13 4.3 L2F Tunnel Establishment 4.3.1 Normal Tunnel Negotiation Sequence 4.3.2 Normal Client Negotiation Sequence 15 4.4 L2F management message types 4.4.1 L2F message type: Invalid 16 4.4.2 L2F_CONF 4.4.3 L2F_OPEN 17 4.4.4 L2F_CLOSE 18 4.4.5 L2F_ECHO 19 4.4.6 L2F_ECHO_RESP 4.5 L2F Message Delivery 20 4.5.1 Sequenced Delivery 4.5.2 Flow Control 4.5.3 Tunnel State Table 4.5.4 Client State Table 21 5.0 Protocol Considerations 5.1 PPP Features 5.2 Termination 22 5.3 Extended Authentication 5.4 MNP4 and Apple Remote Access Protocol 6.0 Acknowledgments 7.0 References 23 8.0 Authors' Addresses 1.0 Introduction The traditional dial-up network service on the Internet is for regis- tered IP addresses only. A new class of virtual dial-up application which allows multiple protocols and unregistered IP addresses is also desired on the Internet. Examples of this class of network applica- tion are support for privately addressed IP, IPX, and AppleTalk dial- up via SLIP/PPP across existing Internet infrastructure. The support of these multiprotocol virtual dial-up applications is of significant benefit to end users and Internet Service providers as it allows the sharing of very large investments in access and core infrastructure and allows local calls to be used. It also allows existing investments in non-IP protocol applications to be supported Valencia expires August 1996 [Page 2] RFC DRAFT January 1996 in a secure manner while still leveraging the access infrastructure of the Internet. It is the purpose of this RFC to identify the issues encountered in integrating multiprotocol dial-up services into an existing Internet Service Provider's Point of Presence (hereafter referred to as ISP and POP, respectively), and to describe the L2F protocol which per- mits the leveraging of existing access protocols. 1.1. Conventions The following language conventions are used in the items of specifi- cation in this document: o MUST, SHALL, or MANDATORY -- This item is an absolute requirement of the specification. o SHOULD or RECOMMEND -- This item should generally be followed for all but exceptional circumstances. o MAY or OPTIONAL -- This item is truly optional and may be followed or ignored according to the needs of the implementor. 2.0 Problem Space Overview In this section we describe in high level terms the scope of the problem that will be explored in more detail in later sections. 2.1 Initial Assumptions We begin by assuming that Internet access is provided by an ISP and that the ISP wishes to offer services other than traditional registered IP address based services to dial-up users of the network. We also assume that the user of such a service wants all of the security facilities that are available to him in a dedicated dial-up configuration. In particular, the end user requires: + End System transparency: Neither the remote end system nor his home site hosts should require any special software to use this service in a secure manner. + Authentication as provided via dial-up PPP CHAP or PAP, or through other dialogs as needed for protocols without authentication (e.g., SLIP). This will include TACACS+ and RADIUS solutions as well as support for smart cards and one-time passwords. The authentication should be manageable by the user independently of the ISP. Valencia expires August 1996 [Page 3] RFC DRAFT January 1996 + Addressing should be as manageable as dedicated dial-up solutions. The address should be assigned by the home site and not the ISP. + Authorization should be managed by the home site as it would in a direct dial-up solution. + Accounting should be performed both by the ISP (for billing purposes) and by the user (for charge-back and auditing). 2.2 Topology Shown below is a generic Internet with Public switched Telephone Network (PSTN) access (i.e., async PPP via modems) and Integrated Services Digital Network (ISDN) access (i.e., synchronous PPP access). Remote users (either async PPP or SLIP, or ISDN) will access the Home LAN as if they were dialed into the Home Gateway, although their physical dial-up is via the ISP Network Access Server. ...----[L]----+---[L]-----... | | [H] | ________|________________________ | | ________|__ ______|________ | | | | | PSTN [R] [R] ISDN | | Cloud | | Cloud [N]__[U] | | Internet | | | | [R] | [N]______[R] |_____________| | | | | | | [U] |________________________________| [H] = Home Gateway [L] = Home LAN(s) [R] = Router [U] = Remote User [N] = ISP Network Access Server ("NAS") Valencia expires August 1996 [Page 4] RFC DRAFT January 1996 2.3 Providing Virtual dial-up Services - a walk-through To motivate the following discussion, this section walks through an example of what might happen when a Virtual dial-up client initiates access. The Remote User initiates a PPP connection to an ISP via either the PSTN or ISDN. The Network Access Server (NAS) accepts the connection and the PPP link is established. The ISP authenticates the end system/user via CHAP or PAP. Only the username field is interpreted to determine whether the user requires a Virtual dial-up service. It is expected that usernames will be structured (e.g. littlewo@cisco.com) or that the ISP maintains a database mapping users to services. In the case of Virtual dial-up, the mapping will name a specific endpoint, the Home Gateway. If a virtual dial-up service is not required, standard access to the Internet may be provided. If no tunnel connection currently exists to the desired Home Gateway, one is initiated. The details of such tunnel creation are outside the scope of this specification; L2F requires only that the tunnel media provide point-to-point connectivity. Obvious examples of such media are UDP, Frame Relay PVC's, or X.25 VC's. Once the tunnel exists, an unused Multiplex ID (hereafter, "MID") is allocated, and a connect indication is sent to notify the Home Gate- way of this new dial-up session. The Home Gateway either accepts the connection, or rejects. Rejection may include a reason indication, which may be displayed to the dial-up user, after which the call should be disconnected. The initial setup notification may include the authentication infor- mation required to allow the Home Gateway to authenticate the user and decide to accept or decline the connection. In the case of CHAP, the set-up packet includes the challenge, username and raw password. For PAP or text dialog (i.e., for SLIP users), it includes username and clear text password. The Home Gateway may choose to use this information to complete its authentication, avoiding an additional cycle of authentication. For PPP, the initial setup notification may also include a copy of the the LCP CONFACKs sent in each direction which completed LCP nego- tiation. The Home Gateway may use this information to initialize its own PPP state (thus avoiding an additional LCP negotiation), or it may choose to initiate a new LCP CONFREQ exchange. Valencia expires August 1996 [Page 5] RFC DRAFT January 1996 If the Home Gateway accepts the connection, it creates a "virtual interface" for SLIP or PPP in a manner analogous to what it would use for a direct-dialed connection. With this "virtual interface" in place, link level frames may now pass over this tunnel in both direc- tions. Frames from the remote user are received at the POP, stripped of any link framing or transparency bytes, encapsulated in L2F, and forwarded over the appropriate tunnel. The Home Gateway accepts these frames, strips L2F, and processes them as normal incoming frames for the appropriate interface and protocol. The "virtual interface" behaves very much like a hardware interface, with the exception that the hardware in this case is physically located at the ISP POP. The other direction behaves analogously, with the Home Gateway encapsulating the packet in L2F, and the POP stripping L2F before transmitting it out the physical interface to the remote user. At this point, the connectivity is a point-to-point PPP or SLIP con- nection whose endpoints are the remote user's networking application on one end and the termination of this connectivity into the Home Gateway's SLIP or PPP support on the other. Because the remote user has become simply another dial-up client of the Home Gateway access server, client connectivity can now be managed using traditional mechanisms with respect to further authorization, protocol access, and filtering. Accounting can be performed at both the NAS as well as the Home Gate- way. This document illustrates some Accounting techniques which are possible using L2F, but the policies surrounding such Accounting are outside the scope of this specification. Because L2F connect notifications for PPP clients contain sufficient information for a Home Gateway to authenticate and initialize its LCP state machine, it is not required that the remote user be queried a second time for CHAP authentication, nor that the client undergo mul- tiple rounds of LCP negotiation and convergence. These techniques are intended to optimize connection setup, and are not intended to deprecate any functions required by the PPP specification. 3.0 Service Model Issues There are several significant differences between the standard Inter- net access service and the Virtual dial-up service with respect to authentication, address allocation, authorization and accounting. The details of the differences between these services and the prob- lems presented by these differences are described below. The mecha- nisms used for Virtual Dial-up service are intended to coexist with more traditional mechanisms; it is intended that an ISP's POP can Valencia expires August 1996 [Page 6] RFC DRAFT January 1996 simultaneously service a ISP clients as well as Virtual dial-up clients. 3.1 Security For the Virtual dial-up service, the ISP pursues authentication only to the extent required to discover the user's apparent identity (and by implication, their desired Home Gateway). As soon as this is determined, a connection to the Home Gateway is initiated with the authentication information gathered by the ISP. The Home Gateway completes the authentication by either accepting the connection, or rejecting it. The Home Gateway must also protect against attempts by third parties to establish tunnels to the Home Gateway. Tunnel establishment involves an ISP-to-Home Gateway authentication phase to protect against such attacks. 3.2 Address Allocation For an Internet service, the user accepts that the IP address may be allocated dynamically from a pool of Service provider addresses. This model often means that the remote user has little or no access to their home network's resources, due to firewalls and other secu- rity policies applied by the home network to accesses from external IP addresses. For the Virtual dial-up service, the Home Gateway can exist behind the home firewall, allocating addresses which are internal (and, in fact, can be RFC1597 addresses, or non-IP addresses). Because L2F tunnels exclusively at the frame level, the actual policies of such address management are irrelevant to correct Virtual dial-up service; for all purposes of PPP or SLIP protocol handling, the dial-in user appears to have connected at the Home Gateway. 3.3 Authentication The authentication of the user occurs in three phases; the first at the ISP, and the second and optional third at the Home gateway. The ISP uses the username to determine that a Virtual dial-up service is required and initiate the tunnel connection to the appropriate Home Gateway. Once a tunnel is established, a new MID is allocated and a session initiated by forwarding the gathered authentication information. The Home Gateway undertakes the second phase by deciding whether or not to accept the connection. The connection indication may include Valencia expires August 1996 [Page 7] RFC DRAFT January 1996 CHAP, PAP, or textual authentication information. Based on this information, the Home Gateway may accept the connection, or may reject it (for instance, it was a PAP request and the user- name/password are found to be incorrect). Once the connection is accepted, the Home Gateway is free to pursue a third phase of authentication at the PPP or SLIP level. These activ- ities are outside the scope of this specification, but might include proprietary PPP extensions, or textual challenges carried via a TCP/IP telnet session. 3.4 Accounting It is a requirement that both the Access gateway and the Home Gateway can provide accounting data and hence both my may count packets, octets and connection start and stop times. Since Virtual dial-up is an access service, accounting of connection attempts (in particular, failed connection attempts) is of signifi- cant interest. The Home Gateway can reject new connections based on the authentication information gathered by the ISP, with correspond- ing logging. For cases where the Home Gateway accepts the connection and then continues with further authentication, the Home Gateway might subsequently disconnect the client. For such scenarios, the disconnection indication back to the ISP may also include a reason. Because the Home Gateway can decline a connection based on the authentication information collected by the ISP, accounting can eas- ily draw a distinction between a series of failed connection attempts and a series of brief successful connections. Lacking this facility, the Home Gateway must always accept connection requests, and would need to exchange a number of PPP packets with the remote system. 4.0 Protocol Definition The protocol definition for Virtual dial-up services requires two areas of standardization: + Encapsulation of PPP packets within L2F: the ISP NAS and the Home gateway require a common understanding of the encapsulation protocol so that SLIP/PPP packets can be successfully transmitted and received across the Internet. + Connection management of L2F and MIDs: the tunnel must be initiated and terminated, as must MIDs within the tunnel. Termination includes diagnostic codes to assist in the diagnosis of problems and to sup- port accounting. Valencia expires August 1996 [Page 8] RFC DRAFT January 1996 While providing these services, the protocol must address the follow- ing required attributes: + Low overhead. The protocol must impose a minimal additional over- head. This requires a compact encapsulation, and a structure for omitting some portions of the encapsulation where their function is not required. + Efficiency. The protocol must be efficient to encapsulate and deencapsulate. + Protocol independence. The protocol must make very few assumptions about the substrate over which L2F packets are carried. 4.1 Encapsulation within L2F 4.1.1 Encapsulation of PPP within L2F The PPP packets may be encapsulated within L2F. The packet encapsu- lated is the packet as it would be transmitted over a physical link. The following are NOT present in the packet: + Flags + Transparency data (ACCM for async, bit stuffing for sync) The following ARE still present: + Address, control, and protocol (unless negotiated away by LCP) + CRC 4.1.2 Encapsulation of SLIP within L2F SLIP is encapsulated within L2F in much the same way as PPP. The transparency characters are removed before encapsulating within L2F, as is the framing. 4.2 L2F Packet Format 4.2.1 Overall Packet Format Valencia expires August 1996 [Page 9] RFC DRAFT January 1996 The entire encapsulated packet has the form: --------------------------------- | | | L2F Header | | | --------------------------------- | | | Payload packet (SLIP/PPP) | | | --------------------------------- | | | Checksum (optional) | | | --------------------------------- 4.2.2 Packet Format An L2F packet has the form: 0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |F|K|0|0|0|0|0|0|0|0|0|0|C| Ver | Protocol | Sequence | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Multiplex ID | Client ID | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Length | Offset (optional) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Key (optional) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ + (payload) | + ..... + + ..... + + ..... + + (payload) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Checksum (optional) | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 4.2.3 Version field The Ver ("Version") field represents the major version of the L2F software creating the packet. It MUST contain the value 001. Valencia expires August 1996 [Page 10] RFC DRAFT January 1996 If any bits are non-zero after bit K but before bit C, an implementa- tion MUST discard the packet and initiate disconnect of the entire tunnel. This would correspond to a packet containing extensions not understood by the receiving end. Handling of the "Key" field must be taken in preference to this processing, to avoid denial-of-service attacks. 4.2.4 Protocol field The Protocol specifies the protocol carried within the L2F packet. Legal values (represented here in hexadecimal) are: Value Type Description 0x00 L2F_ILLEGAL Illegal 0x01 L2F_PROTO L2F management packets 0x02 L2F_PPP PPP tunneled inside L2F 0x03 L2F_SLIP SLIP tunneled inside L2F 4.2.5 Sequence Number The Sequence number starts at 0 for the first L2F packet under a tun- nel. Each subsequent packet is sent with the next increment of the sequence number. The sequence number is thus a free running counter represented modulo 256. For non-L2F management packets, the sequence number is transmitted as 0, does not increment the local sequence counter, and does not affect the processing of received traffic. For L2F management packets, the sequence number is used to protect against duplication of packets, as follows: The receiving side of the tunnel records the sequence number of each valid L2F packet it receives. If a received packet appears to have a value less than or equal to the last received value, the packet MUST be silently discarded. Otherwise, the packet is accepted and the sequence number in the packet recorded as the latest value last received. For purposes of detecting duplication, a received sequence value is considered less than or equal to the last received value if its value lies in the range of the last value and its 127 successor values. For example, if the last received sequence number was 15, then pack- ets with sequence numbers 0 through 15, as well as 144 through 255, would be considered less than or equal to, and would be silently dis- carded. Otherwise it would be accepted. Valencia expires August 1996 [Page 11] RFC DRAFT January 1996 4.2.6 Packet Multiplex ID The Multiplex ID ("MID") identifies a particular connection within the tunnel. Each new connection is assigned a MID currently unused within the tunnel. It is recommended that the MID cycle through the entire 32-bit namespace, to reduce aliasing between previous and cur- rent sessions. The MID with value 0 is special; it is used to communicate the state of the tunnel itself, as distinct from any connection within the tun- nel. 4.2.7 Client ID The Client ID ("CLID") is used to assist endpoints in demultiplexing tunnels when the underlying point-to-point substrate lacks an effi- cient or dependable technique for doing so directly. Using CLID, it is possible to demultiplex multiple tunnels whose packets arrive over the point-to-point media interleaved, without requiring media- specific semantics. When transmitting the L2F_CONF message (described below), the peer's CLID must be communicated via the Assigned_CLID field. This must be a unique non-zero value on the sender's side, which is to be expected in all future non-L2F_CONF packets received. The CLID value from the last valid L2F_CONF message received should be recorded and used as the CLID field value for all subsequent pack- ets sent to the peer. Packets with an unknown Client ID must be silently discarded. For the initial packet sent during tunnel establishment, where no L2F_CONF has yet been received, the CLID field must be 0. Thus, during L2F_CONF each side is told its CLID value. All later packets sent, tagged with this CLID value, serve as a tag which uniquely identifies this peer. 4.2.8 Length Length is the size in octets of the entire packet, including header, all fields present, and payload. Length does not reflect the addi- tion of the checksum, if one is present. The packet should be silently discarded if the received packet is shorter than the indi- cated length. Additional bytes present in the packet beyond the indicated length must be silently ignored. Valencia expires August 1996 [Page 12] RFC DRAFT January 1996 4.2.9 Packet Checksum The Checksum is present if the C bit is present in the header flags. It is a 16-bit CRC as used by PPP/HDLC. Is is applied over the entire packet starting with the first byte of L2F flags, through the last byte of payload data. The checksum is then added as two bytes immediately following the last byte of payload data. 4.2.10 Payload Offset The Offset is present if the F bit is set in the header flags. This field specifies the number of bytes past the header at which the pay- load data is expected to start. If it is 0, or the F bit is not set, the first byte following the last byte of L2F header is the first byte of payload data. It is recommended that data skipped due to the payload offset be ini- tialized to 0's. 4.2.11 Packet Key The Key is the authentication response last given to the peer during tunnel creation (the details of tunnel creation are provided in the next section). It serves as a key during the life of a session to resist attacks based on spoofing. If a packet is received in which the Key does not match the expected value, the packet MUST be silently discarded. 4.3 L2F Tunnel Establishment When the point-to-point link is first initiated between the NAS and the Home Gateway, the endpoints communicate on MID 0 prior to provid- ing general L2F services to clients. This communication is used to verify the presence of L2F on the remote end, and to permit any needed authentication. The protocol for such negotiation is always 1, indicating L2F manage- ment. The message itself is structured as a sequence of single octets indicating an option, followed by zero or more further octets formatted as needed for the option. 4.3.1 Normal Tunnel Negotiation Sequence The establishment sequence is best illustrated by a "typical" connec- tion sequence. Detailed description of each functions follows, along with descriptions of the handling of exceptional conditions. Each packet is described as a source->destination on one line, a Valencia expires August 1996 [Page 13] RFC DRAFT January 1996 description of the L2F packet field contents on the next, and the contents of the packet's body on following lines. The exact encoding of octets will be described later. Note that this example uses the Key option, but does not use the Off- set and Checksum options. The Length field would be present, reflecting the actual length of the packets as encoded as an octet stream. 1. NAS->GW: Proto=L2F, Seq=0, MID=0, CLID=0, Key=0 L2F_CONF Name: NAS_name Challenge: Rnd Assigned_CLID: 22 The NAS decides that a tunnel must be initiated from the NAS to the GW. An L2F packet is sent with the Proto field indicating an L2F management message is contained. Because the tunnel is being initiated, Key is set to 0. The sequence number starts at 0; the MID is 0 to reflect the establishment of the tunnel itself. Since the NAS has not yet received an L2F_CONF, the CLID is set to 0. The body of the packet specifies the claimed name of the NAS, and a challenge random number which GW will use in authenticating itself as a valid tunnel endpoint. Assigned_CLID is generated to be a value not currently assigned out to any other tunnel to any other Home Gateway. 2. GW->NAS: Proto=L2F, Seq=0, MID=0, CLID=22, Key=C(Rnd) L2F_CONF Name: GW_name Challenge: Rnd2 Assigned_CLID: 73 The Home Gateway has processed the previous packet, and sends a response. The protocol continues to be L2F, with a sequence number 0 (each side maintains its own sequence number for transmissions). MID continues to be 0 to reflect tunnel establishment. CLID reflects the Assigned_CLID field of the L2F_CONF received. The Key is a CHAP- style hash of the random number received; each packet hereafter will reflect this calculated value, which serves as a key for the life of the tunnel. The body contains the Home Gateway's name and its own random number Valencia expires August 1996 [Page 14] RFC DRAFT January 1996 challenge. and its own Assigned_CLID for the NAS to place in the CLID field of future packets. The CLID is generated in an analogous manner to that of the NAS. After this, all packets received from the NAS must be tagged with a CLID field containing 73, and all packets sent to the NAS must be tagged with a CLID field containing 22. 3. NAS->GW Proto=L2F, Seq=1, MID=0, CLID=73, Key=C(Rnd2) L2F_OPEN The NAS responds with its Key now set to reflect the shared secret. Like the Home Gateway, the NAS will use this Key for the life of the tunnel. 4. GW->NAS Proto=L2F, Seq=1, MID=0, CLID=22, Key=C(Rnd) L2F_OPEN The Home Gateway provides closure of the key from the NAS. The tun- nel is now available for clients to be established. 4.3.2 Normal Client Negotiation Sequence This section describes the establishment of a Virtual dial-up client on a NAS into a Home Gateway. It assumes a tunnel has been created in the way described in 4.3.1. The client for this example is a PPP client configured for CHAP. Treatment of Checksum, Length, and Offset are as in 4.3.1. 1. NAS->GW Proto=L2F, Seq=2, MID=1, CLID=73, Key=C(Rnd2) L2F_OPEN Authen: CHAP Client: CHAP-name Challenge: Rnd3 Response: The NAS has received a call, tried CHAP with a challenge value of Rnd3, and found that the client responded. The claimed name lead the NAS to believe it was a Virtual dial-up client hosted by the Home Gateway. The next free MID is allocated, and the information associ- ated with the CHAP challenge/response is included in the connect notification. 2. GW->NAS Proto=L2F, Seq=2, MID=1, CLID=22, Key=C(Rnd) L2F_OPEN Valencia expires August 1996 [Page 15] RFC DRAFT January 1996 The Home Gateway, by sending back the L2F_OPEN, accepts the client. 3. NAS->GW Proto=PPP, Seq=0, MID=1, CLID=73, Key=C(Rnd2) 4. GW->NAS Proto=PPP, Seq=0, MID=1, CLID=22, Key=C(Rnd) Traffic is now free to flow in either direction as sent by the remote client or the home site. The contents is uninterpreted data, HDLC in this case. Data traffic, since it is not the L2F protocol, does not use the Seq field, which is set to 0 in non-L2F messages. 4.4 L2F management message types When an L2F packet's Proto field specifies L2F management, the body of the packet is encoded as zero or more options. An option is a single octet "message type", followed by zero or more sub-options. Each sub-option is a single byte sub-option value, and further bytes as appropriate for the sub-option. Options in L2F are: Hex Value Abbreviation Description -------- ------------ ----------- 0x00 Invalid Invalid message 0x01 L2F_CONF Request configuration 0x01 L2F_CONF_TYPE Type of authentication used 0x02 L2F_CONF_NAME Name of peer sending L2F_CONF 0x03 L2F_CONF_CHAL Random number peer challenges with 0x04 L2F_CONF_CLID Assigned_CLID for peer to use 0x02 L2F_OPEN Accept configuration 0x01 L2F_OPEN_CHAP CHAP name received from client 0x02 L2F_OPEN_CHAL Challenge CHAP client received 0x03 L2F_OPEN_RESP CHAP challenge response from client 0x04 L2F_ACK_LCP1 LCP CONFACK accepted from client 0x05 L2F_ACK_LCP2 LCP CONFACK sent to client 0x03 L2F_CLOSE Request disconnect 0x01 L2F_CLOSE_WHY Reason code for close 0x02 L2F_CLOSE_STR ASCII string description 0x04 L2F_ECHO Verify presence of peer 0x05 L2F_ECHO_RESP Respond to L2F_ECHO Valencia expires August 1996 [Page 16] RFC DRAFT January 1996 4.4.1 L2F message type: Invalid If a message is received with this value, or any value higher than the last recognized option value, the packet is considered invalid. The packet MUST be discarded, and an L2F_CLOSE of the entire tunnel MUST be requested. Upon receipt of an L2F_CLOSE, the tunnel itself may be closed. All other received message MUST be discarded. An implementation MAY close the tunnel after an interval of time appro- priate to the characteristics of the tunnel. Invalid sub-option values, even if present under a valid option, must be treated as if the entire message type was invalid. 4.4.2 L2F_CONF The L2F message type is used to establish the tunnel between the NAS and the Home Gateway. MID is always set to 0. The body of such a message starts with the octet 0x01 (L2F_CONF), followed by one more more sub-options. The L2F_CONF_TYPE sub-option must be present. It is encoded as the octet 0x01, followed by a single byte describing the type of authen- tication the NAS exchanged with the client in detecting the client's claimed identification. The authentication types are: 0x01 Textual username/password exchange 0x02 PPP CHAP 0x03 PPP PAP The L2F_CONF_NAME sub-option must be present. It is encoded as the octet value 0x02, followed by an octet specifying a non-zero length, followed by the indicated number of bytes, which are interpreted as the sender's ASCII name. The L2F_CONF_CHAL sub-option must be present. It is encoded as the octet value 0x03, followed by four bytes of challenge value. The challenge value should be generated using whatever techniques provide the highest quality of random numbers available to a given implementation. The L2F_CONF_CLID sub-option must be present. It is encoded as the octet 0x04, followed by four bytes of Assigned_CLID value. The Assigned_CLID value is generated as a non-zero value unique across all tunnels which exist on the sending system. The CLID field is sent as 0 a L2F_CONF packet is received from the peer. After this, the Assigned_CLID value of the last L2F_CONF Valencia expires August 1996 [Page 17] RFC DRAFT January 1996 packet received must be placed in the CLID of all packets being sent. When sent from a NAS to a Home Gateway, the L2F_CONF is the initial packet in the conversation. Key is set to 0, since no challenge has been received yet. When sent from the Home Gateway to the NAS, an L2F_CONF indicates the Home Gateway's recognition of the tunnel creation request. The Home Gateway must provide its name and its own challenge in the message body. Key must be set to the CHAP-style hash of the received chal- lenge bytes. 4.4.3 L2F_OPEN The L2F_OPEN message is used to establish a client connection within a tunnel previously established by L2F_CONF messages. When sent from the NAS to the Home Gateway, it is used to indicate the presence of a new dial-in client. When sent back from the Home Gateway to the NAS, it indicates acceptance of the client. This message starts with the octet 0x02. When sent from the NAS, it may contain further sub- options. When sent from the Home Gateway, it may not contain any options. The L2F_OPEN_CHAP sub-option is encoded as the octet 0x01, followed by an octet specifying the length of the CHAP name received, followed by the indicated number of bytes of CHAP name. The L2F_OPEN_CHAL sub-option is encoded as the octet 0x02, followed by an octet specifying the length of the CHAP challenge sent, fol- lowed by the CHAP challenge itself. The L2F_OPEN_RESP sub-option is encoded as the octet 0x03, followed by an octet specifying the length of the CHAP response sent, followed by the client's response to the CHAP challenge. This message must be treated as invalid if L2F_OPEN_CHAP, L2F_OPEN_CHAL, and L2F_OPEN_RESP do not all appear within the same message. The L2F_ACK_LCP1 and L2F_ACK_LCP2 sub-options are encoded as the octets 0x04 and 0x05 respectively, followed in either case by two octets in network byte order specifying the length of the LCP CONFACK last received from or sent to the client. Following these octets is an exact copy of the CONFACK packet. The Home Gateway may choose to ignore any sub-option of the L2F_OPEN, and accept the connection anyway. The Home Gateway would then have to undertake its own LCP negotiations and authentication. Valencia expires August 1996 [Page 18] RFC DRAFT January 1996 4.4.4 L2F_CLOSE This message is encoded as the byte 0x03. An L2F_CLOSE may be sent by either side of the tunnel at any time. When sent with MID of 0, it indicates the desire to terminate the entire tunnel and all clients within the tunnel. When sent from the Home Gateway in response to an L2F_OPEN, it indicates that the Home Gateway has declined the connection. When sent with a non-zero MID, it indicates the termination of that client within the tunnel. The L2F_CLOSE_WHY sub-option is encoded as the byte 0x01 followed four bytes in network byte order specifying a bit mask of reasons for the disconnection. The bits are encoded as: 0x00000001 Authentication failed 0x00000002 Out of resources 0x00000004 Administrative intervention 0x00000008 User quota exceeded 0x00000010 Protocol error 0x00000020 Unknown user 0x00000040 Incorrect password 0x00000080 PPP configuration incompatible Bits in the mask 0xFF000000 are reserved for per-vendor interpreta- tion. An implementation can choose to not provide status bits even if it detects a condition described by one of these bits. For instance, an implementation may choose to not use 0x00000020 due to security con- siderations, as it can be used to probe user name space. The L2F_CLOSE_STR sub-option is encoded as the byte 0x02, followed by a two-byte length in network byte order, followed by the indicated number of bytes, which are interpreted as descriptive ASCII text associated with the disconnection. This string may be ignored, but could be recorded in a log to provide detailed or auxiliary informa- tion associated with the L2F_CLOSE. 4.4.5 L2F_ECHO Transmission of L2F_ECHO messages is optional. If an implementation transmits L2F_ECHO messages, it must not transmit more than one such request each second. The payload size must be 64 bytes or less in length. It is recommended that at least 5 L2F_ECHO messages be sent without response before an implementation assumes that its peer has terminated. Valencia expires August 1996 [Page 19] RFC DRAFT January 1996 The L2F_ECHO message is encoded as the single byte 0x04. It may be sent by either side once the tunnel is established. MID must be 0. An L2F_ECHO_RESP (documented below) must be sent back in response. 4.4.6 L2F_ECHO_RESP All implementations must respond to L2F_ECHO, using L2F_ECHO_RESP. The received packet must be sent back verbatim, except that the CLID, sequence number, and checksum (if any) must be updated, and the L2F_ECHO message type changed to an L2F_ECHO_RESP. Payload data fol- lowing the 0x04 octet, if any, must be preserved in the response. When an L2F_ECHO_RESP is received, the payload data may be used to associate this response with a previously sent L2F_ECHO, or the packet may be silently discarded. 4.5 L2F Message Delivery L2F is designed to operate over point-to-point unreliable links. It is not designed to provide flow control of the data traffic, nor does it provide reliable deliver of this traffic; each protocol tunnel via L2F is expected to manage flow control and retry itself. Thus, it is only L2F control messages which must be retransmitted; this process is described in this section. 4.5.1 Sequenced delivery All L2F control messages (i.e., those L2F packets with a protocol type of 0x01) are transmitted with a sequence number. The sequence number is a per-L2F tunnel free running counter which is incremented (modulo 256) after each packet is transmitted. It is used to permit the receiving end to detect duplicated or out-of-order packets, and to discard such packets. Section 4.2.5 describes the process in detail. 4.5.2 Flow control L2F control messages are expected to be exchanged lock-step. Thus, per-client activities can not occur until tunnel setup is complete. Neither can one client be serviced until the L2F message exchange is complete for a previous client. Thus, it is expected that rarely--if ever--should a flow control action be required. If the input queue of L2F control messages reaches an objectionable level for an imple- mentation, the implementation may silently discard all messages in the queue to stabilize the situation. Valencia expires August 1996 [Page 20] RFC DRAFT January 1996 4.5.3 Tunnel State table The following enumerates the handling of L2F messages for tunnel cre- ation in state table format. Events name an L2F_ message type (the L2F_ portion of the named message is omitted to permit a more compact table). A start ("*") matches any event not otherwise matched for the named state. State Event Action New State ----- ----- ------ --------- Start1 CONF Send CONF Start2 Start1 * (discard) Start1 Start2 OPEN MID=0 Send OPEN MID=0 Open Start2 CONF Send CONF Start2 Start2 * Send INVALID Start1 Open OPEN MID=0 Send OPEN MID=0 Open Open ECHO CID=0 Send ECHO CID=1 Open Open CLOSE Send CLOSE Close1 Open * Send INVALID Start1 Close1 CLOSE Send CLOSE Close1 Close1 (10 secs) (none) Start1 Close1 * (discard) Close1 4.5.4 Client State table This table is similar to the previous one, but enumerates the states for a client connection within a tunnel in the Open state from 4.5.3. As this sequence addresses clients, MID will be non-zero. State Event Action New State ----- ----- ------ --------- Start1 OPEN Send OPEN Open Start2 * Send INVALID Start1 Open OPEN Send OPEN Open Open CLOSE Send CLOSE Close1 Open (data) Forward Open Open * Send INVALID Start1 Close1 CLOSE Send CLOSE Close1 Close1 (10 secs) (none) Start1 Close1 * (discard) Close1 Valencia expires August 1996 [Page 21] RFC DRAFT January 1996 5. Protocol Considerations Several aspects of operation over L2F, while outside the realm of the protocol description itself, serve to clarify the operation of L2F. 5.1 PPP Features Because L2F in operation carries uninterpreted frames, it permits operation of features without explicit knowledge of these features. For instance, if a PPP session is carried, L2F is simply transporting HDLC frames. The two PPP endpoints can negotiate higher-level fea- tures, such as reliable link, compression, multi-link, or encryption. These features then operate between the two PPP endpoints (the dial- in client on one end, and the Home Gateway on the other), with L2F continuing to simply ship HDLC frames back and forth. For similar reasons, PPP echo requests, NCP configuration negotia- tion, and even termination requests, are all simply tunneled HDLC frames. 5.2 Termination As L2F simply tunnels link-level frames, it does not detect frames like PPP TERMREQ. L2F termination in these scenarios is driven from a protocol endpoint; for instance, if a Home Gateway receives a TERMREQ, its action will be to "hang up" the PPP session. It is the responsibility of the L2F implementation at the Home Gateway to con- vert a "hang up" into an L2F_CLOSE action, which will shut down client's session in the tunnel cleanly. L2F_CLOSE_WHY and L2F_CLOSE_STR may be included to describe the reason for the shut- down. 5.3 Extended Authentication L2F is compatible with both PAP and CHAP protocols. SLIP does not provide authentication within the protocol itself, and thus requires an ASCII exchange of username and password before SLIP is started. L2F is compatible with this mode of operation as well. One-time password cards have become very common. To the extent the NAS can capture and forward the one-time password, L2F operation is compatible with password cards. For the most general solution, an arbitrary request/response exchange must be supported. In an L2F environment, the protocol must be structured so that the NAS can detect the apparent identity of the user and establish a tunnel con- nection to the Home Gateway, where the arbitrary exchange can occur. Valencia expires August 1996 [Page 22] RFC DRAFT January 1996 5.4 MNP4 and Apple Remote Access Protocol L2F appears compatible with Apple's ARAP protocol. Its operation under L2F has not been described simply because this experimental RFC does not have a corresponding implementation of such operation. 6.0 Acknowledgments L2F uses a packet format inspired by GRE. Thanks to Fred Baker for consultation, Dave Carrel for consulting on security aspects, and to Paul Traina for philosophical guidance. 7.0 References [1] Reynolds, J., and J. Postel, "Assigned Numbers", STD 2, RFC 1340, USC/Information Sciences Institute, July 1992. [2] W. Simpson, "The Point-to-Point Protocol (PPP)", RFC 1661, 07/21/1994 [3] S. Hanks, T. Li, D. Farinacci, P. Traina, "Generic Routing Encapsulation (GRE)", RFC 1701, 10/21/1994 [4] J. Romkey, "A Nonstandard for Transmission of IP Datagrams over Serial Lines: SLIP", RFC 1055, June 1988 8.0 Authors' Addresses Tim Kolar Morgan Littlewood Andy Valencia Cisco Systems 170 West Tasman Drive San Jose CA 95134-1706 tkolar@cisco.com littlewo@cisco.com valencia@cisco.com Valencia expires August 1996 [Page 23]